Team:Cornell/project/drylab/modeling/time response

From 2012.igem.org

(Difference between revisions)
(Created page with "{{:Team:Cornell/templates/header}} <!-- paulirish.com/2008/conditional-stylesheets-vs-css-hacks-answer-neither/ --> <!--[if lt IE 7]> <html class="no-js lt-ie9 lt-ie8 lt-ie7" lan...")
 
(76 intermediate revisions not shown)
Line 7: Line 7:
<!--<![endif]-->
<!--<![endif]-->
-
<script type="text/x-mathjax-config">
+
<script type="text/x-mathjax-config">
-
  MathJax.Hub.Config({tex2jax: {inlineMath: [['$','$'], ['\\(','\\)']]}});
+
MathJax.Hub.Config({tex2jax: {inlineMath: [['$','$'], ['\\(','\\)']]}});
-
</script>
+
</script>
-
<script type="text/javascript"
+
<script type="text/javascript"
-
  src="http://cdn.mathjax.org/mathjax/latest/MathJax.js?config=TeX-AMS-MML_HTMLorMML">
+
src="http://cdn.mathjax.org/mathjax/latest/MathJax.js?config=TeX-AMS-MML_HTMLorMML"></script>
-
</script>
+
-
 
+
<div class="row">
<div class="row">
Line 23: Line 21:
<li class="divider"></li>
<li class="divider"></li>
<li>
<li>
-
<a href="https://2012.igem.org/Team:Cornell/project/drylab">Overview</a>
+
<a href="https://2012.igem.org/Team:Cornell/project/drylab">How It Works</a>
-
</li>
+
</li>
-
<li>
+
<li>
-
<a href="https://2012.igem.org/Team:Cornell/project/drylab/specifications">Specifications</a>
+
<a href="https://2012.igem.org/Team:Cornell/project/drylab/functional_requirements">Functional Requirements</a>
-
</li>
+
</li>
-
<li>
+
<li>
-
<a href="https://2012.igem.org/Team:Cornell/project/drylab/design">Design</a>
+
<a href="https://2012.igem.org/Team:Cornell/project/drylab/components">Components</a>
-
</li>
+
</li>
-
<li> Modeling
+
<li>
-
<ul>
+
Modeling
 +
<ul>
<li>
<li>
<a href="https://2012.igem.org/Team:Cornell/project/drylab/modeling/deployment">Deployment</a>
<a href="https://2012.igem.org/Team:Cornell/project/drylab/modeling/deployment">Deployment</a>
-
</li class="active">
+
</li>
-
<li>
+
<li class ="active">
<a href="https://2012.igem.org/Team:Cornell/project/drylab/modeling/time_response">Time Response</a>
<a href="https://2012.igem.org/Team:Cornell/project/drylab/modeling/time_response">Time Response</a>
</li>
</li>
-
</li>
+
</ul>
-
</ul>
+
</li>
-
+
<li>
-
<li>
+
<a href="https://2012.igem.org/Team:Cornell/project/drylab/status">Device Status</a>
-
<a href="https://2012.igem.org/Team:Cornell/project/drylab/gallery">Gallery</a>
+
</li>
-
</li>
+
</ul>
</ul>
</div>
</div>
Line 57: Line 56:
<div class="nine columns">
<div class="nine columns">
-
<h3>Motivation</h3>
+
<h3>How do our sensors respond to varying analyte concentration over time?</h3>
 +
 
 +
To make sense of data collected remotely, we must know the time it takes for our sensor to produce a signal in response to a change in analyte concentration in the sample stream. In other words, we need to know whether current output at any time can be taken to represent analyte concentration <i> at that time </i>, or whether our system has memory. From preliminary data presented for our arsenic sensor in our <a href="https://2012.igem.org/Team:Cornell/project/wetlab/results/currentresponse"> results page</a>, we know the response time of the S.A.F.E. B.E.T. sensor is on the order of 1.5 days. That is, it takes nearly a day and a half for our system to fully respond upon a sudden change in analyte concentration in the reactor.
 +
<br>
 +
<br>
 +
However, knowledge of the characteristic response time is insufficient to predict the dynamic behavior of our sensor when the concentration of analyte is variable in time. In other words, the response time of the sensor can tell us how long it takes our system to respond to a single perturbation in analyte concentration, but further analysis is necessary in the prediction of the sensor’s response to continuously changing analyte concentration. For example, if changes in concentration occur slowly over time, our sensor may produce a signal that perfectly tracks the analyte concentration&#8212;with some lag. On the other hand, if changes occur quickly in relation to the response time of the sensor, we might expect an output corresponding to a time-average of the analyte concentration. Here, we model the current output of our sensor in response to oscillating concentration of analyte in order to predict the time-scales over which the S.A.F.E. B.E.T. sensor filters out fine changes in analyte concentration.
-
The final part of our project is determining where to place our biosensors. We can’t simply throw our biosensors in the river and expect them to detect toxins. Advective and diffusive processes dictate how toxins are transported in a river and if we don’t take these into account we could potentially place our biosensors in a spot where they would never detect any toxins! Why is this true?  Imagine for a moment that there is a river and in this river there is a spot where chemicals are seeping into the river from the soil. As these chemicals seep into the river they are carried downstream by the river through a process called advection. In addition to this, as the chemicals are carried downstream they spread across the width of the river through a process called diffusion. There are several different types of diffusive processes that depend on what scale of diffusion you are looking at. The most pertinent diffusion process for us is turbulent diffusion. If we combine these two processes together the resulting spread of chemicals is shown in Figure 1. As you can see, depending on where we place our biosensor we might never detect any chemicals at all because the chemicals haven’t spread there. If we want to place our biosensors so that they will detect chemicals we need to place them at least a distance $L$ downstream of the chemical source. However, if we place our biosensors too far downstream of the chemical source then the concentration of chemicals will be too dilute for our device to measure. So its best to place our biosensor a distance $L$ away from the chemical source.
 
</div>
</div>
<div class="three columns">
<div class="three columns">
-
<img src="https://static.igem.org/mediawiki/2012/8/8f/Map_Small.jpeg">
+
<img src="https://static.igem.org/mediawiki/2012/a/a7/Cornell_TR_River.jpg">
</div>
</div>
</div>
</div>
<div class="row" id="item2">
<div class="row" id="item2">
-
<div class="three columns">
+
<div class="four columns" style="text-align:center;">
-
<img src="https://static.igem.org/mediawiki/2012/8/8f/Map_Small.jpeg">
+
<a href="https://static.igem.org/mediawiki/2012/6/6e/Cornell_RT_Reactor.png" rel="lightbox"> <img class="inline" src="https://static.igem.org/mediawiki/2012/2/2f/Cornell_RT_Reactor_SQ.png">Click to Enlarge</a>
 +
 
</div>
</div>
-
<div class="nine columns">
+
<div class="eight columns">
-
<h3>Theory</h3>
+
<h3>Problem Setup</h3>
-
So for a given river width $W$, how do we find the distance $L$ to place our biosensor? One way to do that would be to go to the river in question, measure its height, width, length and volumetric flow rate and numerically solve the advection diffusion equation. However the Alberta Oil Sands are roughly 4,000 km away from Cornell so it was impossible for us to go there and measure the rivers. Instead we decided to use dimensional analysis to estimate what the value of $L$ would be for different rivers. To do this we first defined a time scale $\tau$ which is the time scale for advection. $\tau$ can be written as
+
As seen in the figure to the left, the S.A.F.E. B.E.T. sensor is a continuous flow reactor, with the feed rate at the inlet equal to the effluent flow rate&#8212;at a value of $F$ = 18 mL/min. The reactor has constant volume of 120 mL.
 +
<br><br>
 +
We want to know how our system responds to oscillating analyte concentration in the feed stream&#8212;<i>i.e.</i>, the river in which the sensor is deployed. To this end, we will model time varying analyte concentration  $([A](t))$ as a sine wave. The frequency of this wave will be the parameter of primary interest, since we are particularly interested in determining the timescales over which the S.A.F.E. B.E.T. sensor produces a signal that tracks fine changes in analyte concentration.
 +
<br><br>
 +
However, before we can determine our sensor's current response to an input function of varying river analyte concentration, we must first solve for the concentration of analyte <i>in the reactor</i> as a function of time $(c(t))$. Therefore, our solution will consist of two coupled differential equations&#8212;one which may be solved analytically for $c(t)$ and one which may be numerically solved for the current response of the S.A.F.E. B.E.T. sensor&#8212;as described below.
-
$$\tau = \frac{L}{u}$$
 
-
where $u$ is the average velocity of the river and $L$ is the length scale for advection. Similarly we defined a time scale for turbulent diffusion which can be written as
+
</div>
-
$$\tau=  \frac{l^2}{D}$$
+
</div>
-
where $l$ is the length scale for turbulent diffusion and $D$ is the turbulent diffusion coefficient. Since advection and diffusion are happening at the same time we can set both time scales equal to each other and solve for $L$ which yields
+
<div class="row" id="item3">
 +
<div class="nine columns">
-
$$L=  \frac{ul^2}{D}$$
+
<h3>Solution</h3>
 +
<h5>Concentration of Analyte In Reactor</h5>
 +
First, we need to solve for the concentration of analyte in the reactor as a function of time $(c(t))$ given an oscillating input of analyte $([A](t))$&#8212;with arbitrary frequency and amplitude&#8212;in the river. By performing a mass balance around the reactor (and ignoring any contribution from generation or consumption), we may write the following differential equation:
-
The average velocity $u$ can be rewritten as
+
$$\frac{dc}{dt} = [A](t)\cdot\frac{F}{V}-c\cdot\frac{F}{V}=D([A](t)-c),$$
-
$$u = \frac{Q}{wh}$$
+
$$ \mathrm{where} \ \ [A](t) = A_0\cdot\sin(2 \pi ft)+A_0,$$
 +
and the dilution rate, $D$, equals 3.6 day$^{-1}$ for our system. Using $e^{Dt}$ as an integrating factor, we find
-
where $Q$ is the volumetric flow rate of the river, $w$ is the width of the river and $h$ is the depth of the river. Next, the turbulent diffusion coefficient can be rewritten as
+
$$c(t) = \frac{(A0 (D^2 \sin(2 \pi f t)+D^2-2 \pi D f \cos(2 \pi f t)+4 \pi^2 f^2))}{(D^2+4 \pi^2 f^2)}+k_1 e^{-D t}$$
-
$$ D=kh \sqrt{ghs}$$
+
$$ \mathrm{where} \ \ k_1 = - \frac{(A_0 (D^2-2 \pi D f+4 \pi^2))}{(D^2+4 \pi^2 f^2)} \ \  \mathrm{such\ that} \ c(0) =0 $$
-
where $s$ is the slope of the river, $g$ is the acceleration due to gravity and $k$ is an experimentally determined constant. This constant differs for turbulent diffusion across a river (horizontal) and turbulent diffusion within a river (vertical). Since we want to place our biosensor at the point where the chemical is completely mixed in the river, we need to take into account both horizontal and vertical turbulent diffusion. Putting both of these definitions into our equation for $L$ gives us
+
<br>
 +
<br>
 +
<h5>Current Response</h5>
 +
Now that we have an analytical solution for the concentration of analyte in the reactor over time, we wish to model the current response given the input function $c(t)$. To accomplish this, we model the time rate of change of current of our arsenic sensor using a Hill function [1] with a cooperativity coefficient of unity&#8212;fitting data presented in Fig. 3 of our <a href="https://2012.igem.org/Team:Cornell/project/wetlab/results/currentresponse">Current Response </a>characterization page&#8212;and lumping all transcriptional, translational, and post-translational processes:
-
$$L=  \frac{Ql^2}{kwh^2\sqrt{ghs}}$$
+
$$\frac{dI}{dt} = \frac{\beta_1 \cdot c(t)}{c_{1/2}+c(t)}+\beta_0 - \alpha I, $$
-
Finally, we need to define what $k$ and $l$ are. If we want to know how far downstream the chemicals will be mixed horizontally across the river ($L_Y$) then $l$ = $w$ and $k$ = 0.6 [1]. If we want to know how far downstream the chemicals will be mixed vertically in the river ($L_Z$) then $l$ = $h$ and $k$ = 0.067 [1]. The larger of these $L$ values will determine where we need to place our biosensor for a giver river. This gives us the following two equations
+
where $c(t)$ is as defined above; $\beta_1$ is 1.4 $\mu$A/day, $\beta_0$ is 2.8 $\mu$A/day, $c_{1/2}$, the half-saturating analyte concentration, is 100 $\mu$M, $\alpha$ = 0.69 day$^{-1}$; and the saturating current output is $(\beta_1 + \beta_0)/\alpha$.
 +
<br>
 +
<br>
 +
Using this ordinary differential equation, we may numerically solve for current as a function of time&#8212;given the input function derived above. The results of such numerical solutions for input functions with various frequencies&#8212;<a href="https://static.igem.org/mediawiki/2012/6/67/ShewieTime.m" target="_blank">performed </a>using the differential equation solver ode45 in MATLAB&#8212;are shown below.
-
$$L_Y= \frac{Qw}{0.6h^2\sqrt{ghs}}$$
+
</div>
-
$$L_Z= \frac{Q}{0.067w\sqrt{ghs}}$$
+
<div class="three columns">
 +
<img src="https://static.igem.org/mediawiki/2012/d/d0/Cornell_mathhhh.jpg">
 +
</div>
 +
</div>
 +
 
 +
<div class="row" id="item4">
 +
<div class="three columns">
 +
<img src="https://static.igem.org/mediawiki/2012/0/07/Cornell_MATLAB.png">
 +
</div>
 +
<div class="nine columns">
 +
<h3>Results: Time-Averaged Output for Rapidly Oscillating Analyte Concentrations</h3>
 +
As seen below in Fig. 1, variations in analyte concentration on the order of 90 days can be detected by the S.A.F.E. B.E.T. sensor without time-averaging. In other words, the signal produced by the sensor tracks the changes in analyte without disruption from the ‘memory’ of the system. This is important, since we know our sensor can reliably report analyte concentration at any given time once we account for the simple time lag of the response time&#8212;which is negligible for variations on timescales of weeks or greater.
 +
<br><br>
 +
<img class="inline" src="https://static.igem.org/mediawiki/2012/a/a7/Cornell_TR1.png">
 +
<h6>Fig. 1. A sinusoidally oscillating concentration of analyte (with period, T, of 90 days) in the feed stream (green) is plotted alongside the modeled sensor output (blue). For analyte changes on this timescale, current output traverses the entire range of the outputs from basal current of 4 $\mu$A to saturating current just under 6 $\mu$A&#8212; giving a fine-tuned output corresponding to analyte concentrations within the dynamic range. </h6>
 +
 
 +
<br>
 +
<br><br>
 +
However, for variations in analyte that occur on timescales of less than a month, the ‘memory’ of the system begins to interfere with our sensor’s ability to track fine changes in analyte concentration. As seen in Fig. 2 below, when oscillations in analyte concentration occur on the order of days, the response of our sensor approaches a constant output that accurately represents the time-average of the oscillating analyte concentration. Even though S.A.F.E. B.E.T. cannot track fine changes on these timescales, the ability to sense the time-averaged value represents a significant improvement over current detection methods that rely on ‘spot’ sampling&#8212;which often misrepresents the true analyte concentration. 
 +
<br><br>
 +
 
 +
<img class="inline" src="https://static.igem.org/mediawiki/2012/2/29/Cornell_TR2.png">
 +
<h6>Fig. 2. Modeled current output of the S.A.F.E. B.E.T. sensor is plotted as a function of time for input functions corresponding to varying river analyte concentrations&#8212;with oscillatory periods of 90 days (blue), 10 days (green) and 1 day (black). As in Fig. 1, for analyte changes on the order of 90 days, the sensor produces a fine-tuned output corresponding to analyte concentrations within its dynamic range. As the frequency of analyte variation increases, the current response approaches that expected from the time-averaged concentration (pink dash).        </h6>
</div>
</div>
Line 105: Line 144:
</div>
</div>
-
 
-
 
-
 
<div class="row last-ele">
<div class="row last-ele">
<div class="twelve columns">
<div class="twelve columns">
<h3>References</h3>
<h3>References</h3>
-
[1] https://ceprofs.civil.tamu.edu/ssocolofsky/cven489/downloads/book/ch3.pdf
+
[1] Hill, A. (1910) The possible effects of the aggregation of the molecules of haemoglobin on its dissociation curves. J Physiol 40: 4–7.
-
<br><br>
+
-
[2] http://www.environment.alberta.ca/apps/basins/Map.aspx?Basin=2&DataType=1
+
-
<br><br>
+
-
[3] http://www.ramp-alberta.org/data/ClimateHydrology/hydrology/default.aspx
+
</div>
</div>
</div>
</div>
-
 
-
 
-
 
-
 
</div>
</div>
Line 129: Line 157:
<script src="https://2012.igem.org/Team:Cornell/javascripts/foundation.min?action=raw&amp;ctype=text/javascript"></script>
<script src="https://2012.igem.org/Team:Cornell/javascripts/foundation.min?action=raw&amp;ctype=text/javascript"></script>
<script src="https://2012.igem.org/Team:Cornell/javascripts/app?action=raw&amp;ctype=text/javascript"></script>
<script src="https://2012.igem.org/Team:Cornell/javascripts/app?action=raw&amp;ctype=text/javascript"></script>
 +
<script src="https://2012.igem.org/Team:Cornell/javascripts/lightbox?action=raw&amp;ctype=text/javascript"></script>
<script type="text/javascript">
<script type="text/javascript">
$(window).load(function() {
$(window).load(function() {

Latest revision as of 07:35, 10 November 2012

Time Response

How do our sensors respond to varying analyte concentration over time?

To make sense of data collected remotely, we must know the time it takes for our sensor to produce a signal in response to a change in analyte concentration in the sample stream. In other words, we need to know whether current output at any time can be taken to represent analyte concentration at that time , or whether our system has memory. From preliminary data presented for our arsenic sensor in our results page, we know the response time of the S.A.F.E. B.E.T. sensor is on the order of 1.5 days. That is, it takes nearly a day and a half for our system to fully respond upon a sudden change in analyte concentration in the reactor.

However, knowledge of the characteristic response time is insufficient to predict the dynamic behavior of our sensor when the concentration of analyte is variable in time. In other words, the response time of the sensor can tell us how long it takes our system to respond to a single perturbation in analyte concentration, but further analysis is necessary in the prediction of the sensor’s response to continuously changing analyte concentration. For example, if changes in concentration occur slowly over time, our sensor may produce a signal that perfectly tracks the analyte concentration—with some lag. On the other hand, if changes occur quickly in relation to the response time of the sensor, we might expect an output corresponding to a time-average of the analyte concentration. Here, we model the current output of our sensor in response to oscillating concentration of analyte in order to predict the time-scales over which the S.A.F.E. B.E.T. sensor filters out fine changes in analyte concentration.

Problem Setup

As seen in the figure to the left, the S.A.F.E. B.E.T. sensor is a continuous flow reactor, with the feed rate at the inlet equal to the effluent flow rate—at a value of $F$ = 18 mL/min. The reactor has constant volume of 120 mL.

We want to know how our system responds to oscillating analyte concentration in the feed stream—i.e., the river in which the sensor is deployed. To this end, we will model time varying analyte concentration $([A](t))$ as a sine wave. The frequency of this wave will be the parameter of primary interest, since we are particularly interested in determining the timescales over which the S.A.F.E. B.E.T. sensor produces a signal that tracks fine changes in analyte concentration.

However, before we can determine our sensor's current response to an input function of varying river analyte concentration, we must first solve for the concentration of analyte in the reactor as a function of time $(c(t))$. Therefore, our solution will consist of two coupled differential equations—one which may be solved analytically for $c(t)$ and one which may be numerically solved for the current response of the S.A.F.E. B.E.T. sensor—as described below.

Solution

Concentration of Analyte In Reactor
First, we need to solve for the concentration of analyte in the reactor as a function of time $(c(t))$ given an oscillating input of analyte $([A](t))$—with arbitrary frequency and amplitude—in the river. By performing a mass balance around the reactor (and ignoring any contribution from generation or consumption), we may write the following differential equation: $$\frac{dc}{dt} = [A](t)\cdot\frac{F}{V}-c\cdot\frac{F}{V}=D([A](t)-c),$$ $$ \mathrm{where} \ \ [A](t) = A_0\cdot\sin(2 \pi ft)+A_0,$$ and the dilution rate, $D$, equals 3.6 day$^{-1}$ for our system. Using $e^{Dt}$ as an integrating factor, we find $$c(t) = \frac{(A0 (D^2 \sin(2 \pi f t)+D^2-2 \pi D f \cos(2 \pi f t)+4 \pi^2 f^2))}{(D^2+4 \pi^2 f^2)}+k_1 e^{-D t}$$ $$ \mathrm{where} \ \ k_1 = - \frac{(A_0 (D^2-2 \pi D f+4 \pi^2))}{(D^2+4 \pi^2 f^2)} \ \ \mathrm{such\ that} \ c(0) =0 $$

Current Response
Now that we have an analytical solution for the concentration of analyte in the reactor over time, we wish to model the current response given the input function $c(t)$. To accomplish this, we model the time rate of change of current of our arsenic sensor using a Hill function [1] with a cooperativity coefficient of unity—fitting data presented in Fig. 3 of our Current Response characterization page—and lumping all transcriptional, translational, and post-translational processes: $$\frac{dI}{dt} = \frac{\beta_1 \cdot c(t)}{c_{1/2}+c(t)}+\beta_0 - \alpha I, $$ where $c(t)$ is as defined above; $\beta_1$ is 1.4 $\mu$A/day, $\beta_0$ is 2.8 $\mu$A/day, $c_{1/2}$, the half-saturating analyte concentration, is 100 $\mu$M, $\alpha$ = 0.69 day$^{-1}$; and the saturating current output is $(\beta_1 + \beta_0)/\alpha$.

Using this ordinary differential equation, we may numerically solve for current as a function of time—given the input function derived above. The results of such numerical solutions for input functions with various frequencies—performed using the differential equation solver ode45 in MATLAB—are shown below.

Results: Time-Averaged Output for Rapidly Oscillating Analyte Concentrations

As seen below in Fig. 1, variations in analyte concentration on the order of 90 days can be detected by the S.A.F.E. B.E.T. sensor without time-averaging. In other words, the signal produced by the sensor tracks the changes in analyte without disruption from the ‘memory’ of the system. This is important, since we know our sensor can reliably report analyte concentration at any given time once we account for the simple time lag of the response time—which is negligible for variations on timescales of weeks or greater.

Fig. 1. A sinusoidally oscillating concentration of analyte (with period, T, of 90 days) in the feed stream (green) is plotted alongside the modeled sensor output (blue). For analyte changes on this timescale, current output traverses the entire range of the outputs from basal current of 4 $\mu$A to saturating current just under 6 $\mu$A— giving a fine-tuned output corresponding to analyte concentrations within the dynamic range.



However, for variations in analyte that occur on timescales of less than a month, the ‘memory’ of the system begins to interfere with our sensor’s ability to track fine changes in analyte concentration. As seen in Fig. 2 below, when oscillations in analyte concentration occur on the order of days, the response of our sensor approaches a constant output that accurately represents the time-average of the oscillating analyte concentration. Even though S.A.F.E. B.E.T. cannot track fine changes on these timescales, the ability to sense the time-averaged value represents a significant improvement over current detection methods that rely on ‘spot’ sampling—which often misrepresents the true analyte concentration.

Fig. 2. Modeled current output of the S.A.F.E. B.E.T. sensor is plotted as a function of time for input functions corresponding to varying river analyte concentrations—with oscillatory periods of 90 days (blue), 10 days (green) and 1 day (black). As in Fig. 1, for analyte changes on the order of 90 days, the sensor produces a fine-tuned output corresponding to analyte concentrations within its dynamic range. As the frequency of analyte variation increases, the current response approaches that expected from the time-averaged concentration (pink dash).

References

[1] Hill, A. (1910) The possible effects of the aggregation of the molecules of haemoglobin on its dissociation curves. J Physiol 40: 4–7.